the following code:
{$mode objfpc}
var
  a: Integer;
begin
  a := a + 1;
  asm nop end;
end.

generates (under -O2):

movl U_P$PROGRAM_A,%eax
incl %eax
movl %eax,U_P$PROGRAM_A
nop

instead of a single "incl".
removing either "asm" statement OR $mode directive reinstates the optimization (although in the latter case with "incw", of course).

This obviously never matters in practice, but may be a symptom of some more serious problem. Should I file a bug report?

FPC doesn't do any optimization in procedures containing inline assembler.

Florian Klämpfl wrote

FPC doesn't do any optimization in procedures containing inline assembler.

It does. But also assumes that all registers are used in assembler blocks, unless the programmer specifies otherwise. If you change the assembler block into

   asm nop end [];

then the statement will optimized. Otherwise the compiler assumes that the contents of eax may still be used inside the assembler block and hence cannot remove its use.

The fact that the optimization is performed in case the used register is ax instead of eax suggests a small bug.

Jonas

2011/10/3 Jonas Maebe :

I see, thanks.

It does. But also assumes that all registers are used in assembler blocks, unless the programmer specifies otherwise.

But a:=a+1; does not imply any register usage, and the variable is a
global one?

DoDi

Oops, I thought you changed this a while ago :)